Five ML Concepts - #1

5 machine learning concepts. Under 30 seconds each.

Resource	Link
Papers	Links in References section
Video	Five ML Concepts #1

References

Concept	Reference
Backprop	Learning representations by back-propagating errors (Rumelhart, Hinton, Williams 1986)
Transformer	Attention Is All You Need (Vaswani et al. 2017)
Mamba	Mamba: Linear-Time Sequence Modeling (Gu & Dao 2023)
Hallucination	Survey of Hallucination in NLG (Ji et al. 2023)
Embedding	Word2Vec (Mikolov et al. 2013)

Today’s Five

1. Backpropagation

Back propagation of errors. It’s how neural networks learn—flowing error backward through the network to adjust each weight.

Without it, modern deep learning wouldn’t be practical.

Think of it like retracing your steps to see which earlier choices caused the mistake.

2. Transformer

The architecture behind GPT, Claude, and most modern language models. Instead of processing words one at a time, transformers use attention to weigh relationships between all tokens.

This enables parallel training and rich context awareness.

Like understanding a sentence by seeing how every word relates to every other.

3. Mamba (State Space Models)

A newer alternative to transformers that processes sequences in linear time instead of quadratic.

This allows scaling to very long documents with much lower memory use.

Like a smart conveyor belt that carries forward only what matters.

4. Hallucination

When a model generates confident-sounding nonsense. It happens because language models predict plausible next words, not true facts.

They optimize for likelihood, not correctness.

Like a student who writes confidently without verifying sources.

5. Embedding

Turning words, images, or concepts into vectors of numbers. Similar meanings end up close together in this space.

This lets math capture semantic relationships.

Think of it as a coordinate system for meaning.

Quick Reference

Concept	One-liner
Backprop	Learn by flowing error backward
Transformer	Attention over all tokens at once
Mamba	Linear-time sequence modeling
Hallucination	Confident nonsense from likelihood optimization
Embedding	Meaning as coordinates in vector space

---

Short, accurate ML explainers. Follow for more.